import os
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Rescaling, Conv2D, MaxPooling2D, Dense, Flatten, Dropout, BatchNormalization
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.optimizers import Adam
from sklearn.metrics import confusion_matrix, precision_score, recall_score, accuracy_score, f1_score
# location path of the datasets
train_dir = "/Users/preslav/Downloads/cw_cop528/imageset/train"
test_dir = "/Users/preslav/Downloads/cw_cop528/imageset/val"
# setting a common standard for the pixel values, to fall in
# setting a validation and training split, alongside augmentation
# details about the train dataset
train_data = ImageDataGenerator(rescale=1./255,
rotation_range=40,
shear_range=0.2,
zoom_range=0.3,
horizontal_flip=True,
fill_mode="nearest",
width_shift_range=0.3,
height_shift_range=0.3,
validation_split=0.2)
val_data = ImageDataGenerator(rescale=1/255,
validation_split=0.2)
test_data = ImageDataGenerator(rescale=1./255)
# importing the data batches and setting their properties
train_batches = train_data.flow_from_directory(directory = train_dir,
target_size = (224, 224),
subset = "training",
batch_size = 32,
seed = 2)
validation_batches = val_data.flow_from_directory(directory = train_dir,
target_size = (224, 224),
subset = "validation",
batch_size = 32,
seed = 2)
test_batches = test_data.flow_from_directory(directory = test_dir,
target_size = (224, 224),
batch_size = 32,
shuffle = False)
Found 7578 images belonging to 10 classes. Found 1891 images belonging to 10 classes. Found 3925 images belonging to 10 classes.
# import of the class labels names and their total number
class_names = list(train_batches.class_indices.keys())
num_classes = len(class_names)
print(class_names)
print(num_classes)
['building', 'dog', 'fish', 'gas_station', 'golf', 'musician', 'parachute', 'radio', 'saw', 'vehicle'] 10
# importing a batch of images and labels
img, lbl = next(train_batches)
# plotting 9 images and their respective class labels
plt.figure(figsize = (12, 12))
for i in range(9):
class_index = np.argmax(lbl[i])
plt.subplot(3, 3, i + 1)
plt.imshow(img[i])
plt.title(class_names[class_index])
plt.axis("off")
plt.tight_layout()
plt.show()
# setting the model's architecture
model_adapted_augmented = Sequential([
Conv2D(16, (3,3), 1, activation="relu"),
BatchNormalization(),
MaxPooling2D(),
Conv2D(32, (3,3), 1, activation="relu"),
BatchNormalization(),
Dropout(0.25),
Conv2D(32, (3,3), 1, activation="relu"),
BatchNormalization(),
Conv2D(32, (3,3), 1, activation="relu"),
BatchNormalization(),
MaxPooling2D(),
Conv2D(32, (3,3), 1, activation="relu"),
BatchNormalization(),
Dropout(0.25),
Conv2D(32, (3,3), 1, activation="relu"),
BatchNormalization(),
MaxPooling2D(),
Flatten(),
Dense(256, activation="relu"),
BatchNormalization(),
Dropout(0.5),
Dense(num_classes, activation="softmax")
])
Metal device set to: Apple M2
2023-03-17 11:43:26.503498: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support. 2023-03-17 11:43:26.503609: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:271] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)
# setting the model's loss function, gradient descnet optimizer and evaluation metrics
model_adapted_augmented.compile(optimizer = "adam", loss = "categorical_crossentropy", metrics = ["accuracy"])
# performing training of the model with the training batches and validaiton batches
epochs = 20
history_adapted_augmented = model_adapted_augmented.fit(train_batches,
validation_data = validation_batches,
epochs = epochs)
Epoch 1/20
2023-03-17 11:43:27.095353: W tensorflow/core/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz 2023-03-17 11:43:27.564403: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.
237/237 [==============================] - ETA: 0s - loss: 2.2734 - accuracy: 0.2666
2023-03-17 11:44:15.094136: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.
237/237 [==============================] - 52s 214ms/step - loss: 2.2734 - accuracy: 0.2666 - val_loss: 7.9458 - val_accuracy: 0.1111 Epoch 2/20 237/237 [==============================] - 51s 215ms/step - loss: 1.9513 - accuracy: 0.3394 - val_loss: 18.5200 - val_accuracy: 0.1512 Epoch 3/20 237/237 [==============================] - 51s 214ms/step - loss: 1.8802 - accuracy: 0.3580 - val_loss: 2.0102 - val_accuracy: 0.3887 Epoch 4/20 237/237 [==============================] - 50s 212ms/step - loss: 1.7232 - accuracy: 0.4212 - val_loss: 2.5158 - val_accuracy: 0.3210 Epoch 5/20 237/237 [==============================] - 50s 212ms/step - loss: 1.6860 - accuracy: 0.4274 - val_loss: 1.6564 - val_accuracy: 0.4870 Epoch 6/20 237/237 [==============================] - 50s 212ms/step - loss: 1.5740 - accuracy: 0.4702 - val_loss: 1.5357 - val_accuracy: 0.5331 Epoch 7/20 237/237 [==============================] - 51s 213ms/step - loss: 1.5051 - accuracy: 0.4950 - val_loss: 2.0012 - val_accuracy: 0.4601 Epoch 8/20 237/237 [==============================] - 51s 213ms/step - loss: 1.4504 - accuracy: 0.5182 - val_loss: 1.3038 - val_accuracy: 0.5838 Epoch 9/20 237/237 [==============================] - 51s 213ms/step - loss: 1.3965 - accuracy: 0.5376 - val_loss: 2.9507 - val_accuracy: 0.4812 Epoch 10/20 237/237 [==============================] - 50s 212ms/step - loss: 1.3798 - accuracy: 0.5383 - val_loss: 2.5158 - val_accuracy: 0.3845 Epoch 11/20 237/237 [==============================] - 50s 213ms/step - loss: 1.3507 - accuracy: 0.5414 - val_loss: 1.4932 - val_accuracy: 0.5056 Epoch 12/20 237/237 [==============================] - 51s 212ms/step - loss: 1.3166 - accuracy: 0.5652 - val_loss: 2.1749 - val_accuracy: 0.5056 Epoch 13/20 237/237 [==============================] - 50s 212ms/step - loss: 1.2959 - accuracy: 0.5681 - val_loss: 3.2859 - val_accuracy: 0.3982 Epoch 14/20 237/237 [==============================] - 50s 212ms/step - loss: 1.2866 - accuracy: 0.5744 - val_loss: 1.9545 - val_accuracy: 0.5278 Epoch 15/20 237/237 [==============================] - 51s 213ms/step - loss: 1.2399 - accuracy: 0.5900 - val_loss: 2.1397 - val_accuracy: 0.4246 Epoch 16/20 237/237 [==============================] - 51s 213ms/step - loss: 1.2264 - accuracy: 0.5897 - val_loss: 1.4010 - val_accuracy: 0.5690 Epoch 17/20 237/237 [==============================] - 50s 212ms/step - loss: 1.2528 - accuracy: 0.5830 - val_loss: 2.4072 - val_accuracy: 0.4146 Epoch 18/20 237/237 [==============================] - 51s 213ms/step - loss: 1.2449 - accuracy: 0.5900 - val_loss: 1.3345 - val_accuracy: 0.5785 Epoch 19/20 237/237 [==============================] - 50s 212ms/step - loss: 1.2054 - accuracy: 0.6002 - val_loss: 1.5779 - val_accuracy: 0.6621 Epoch 20/20 237/237 [==============================] - 51s 214ms/step - loss: 1.1713 - accuracy: 0.6134 - val_loss: 1.8299 - val_accuracy: 0.5193
# getting model's summary
model_adapted_augmented.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, None, None, 16) 448
batch_normalization (BatchN (None, None, None, 16) 64
ormalization)
max_pooling2d (MaxPooling2D (None, None, None, 16) 0
)
conv2d_1 (Conv2D) (None, None, None, 32) 4640
batch_normalization_1 (Batc (None, None, None, 32) 128
hNormalization)
dropout (Dropout) (None, None, None, 32) 0
conv2d_2 (Conv2D) (None, None, None, 32) 9248
batch_normalization_2 (Batc (None, None, None, 32) 128
hNormalization)
conv2d_3 (Conv2D) (None, None, None, 32) 9248
batch_normalization_3 (Batc (None, None, None, 32) 128
hNormalization)
max_pooling2d_1 (MaxPooling (None, None, None, 32) 0
2D)
conv2d_4 (Conv2D) (None, None, None, 32) 9248
batch_normalization_4 (Batc (None, None, None, 32) 128
hNormalization)
dropout_1 (Dropout) (None, None, None, 32) 0
conv2d_5 (Conv2D) (None, None, None, 32) 9248
batch_normalization_5 (Batc (None, None, None, 32) 128
hNormalization)
max_pooling2d_2 (MaxPooling (None, None, None, 32) 0
2D)
flatten (Flatten) (None, None) 0
dense (Dense) (None, 256) 4718848
batch_normalization_6 (Batc (None, 256) 1024
hNormalization)
dropout_2 (Dropout) (None, 256) 0
dense_1 (Dense) (None, 10) 2570
=================================================================
Total params: 4,765,226
Trainable params: 4,764,362
Non-trainable params: 864
_________________________________________________________________
# Graphical evaluation of training performance
acc = history_adapted_augmented.history['accuracy']
val_acc = history_adapted_augmented.history['val_accuracy']
loss = history_adapted_augmented.history['loss']
val_loss = history_adapted_augmented.history['val_loss']
epochs_range = range(epochs)
plt.figure(figsize=(11, 8))
plt.subplots_adjust(hspace = .3)
plt.subplot(2, 1, 1)
plt.plot(epochs_range, acc, label = 'Training Accuracy', color = "orange")
plt.plot(epochs_range, val_acc, label = 'Validation Accuracy', color = "blue")
plt.legend(loc = 'best')
plt.xlabel('Epochs')
plt.title('Training and Validation Accuracy', size = 13)
plt.subplot(2, 1, 2)
plt.plot(epochs_range, loss, label = 'Training Loss', color = "orange")
plt.plot(epochs_range, val_loss, label = 'Validation Loss', color = "blue")
plt.legend(loc = 'best')
plt.title('Training and Validation Loss', size = 13)
plt.ylim(0, 3.5)
plt.xlabel('Epochs')
plt.suptitle("Base Model's Updated Architecture with Data Augmentation", size=15)
plt.show()
# Test loss and accuracy measurments
test_loss, test_acc = model_adapted_augmented.evaluate(test_batches)
print('Test loss:', test_loss)
print('Test accuracy:', test_acc)
123/123 [==============================] - 7s 60ms/step - loss: 1.7517 - accuracy: 0.5131 Test loss: 1.7516899108886719 Test accuracy: 0.5131210684776306
# getting prediction labales by running the softmax results in argmax
test_labels = test_batches.classes
y_pred = model_adapted_augmented.predict(test_batches)
predicted_lables = np.argmax(y_pred, axis = 1)
cm = confusion_matrix(test_labels, predicted_lables)
1/123 [..............................] - ETA: 25s
2023-03-17 12:00:27.553762: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.
123/123 [==============================] - 7s 58ms/step
# dataframe containing the confussion matrix
cfm = pd.DataFrame(cm, index = class_names, columns = class_names)
# plotting the conffusion matrix
sns.heatmap(cfm, annot=True, fmt='d', cmap='Purples')
plt.xlabel('Predicted Label')
plt.ylabel('True Label')
plt.xticks(rotation=78)
plt.title("Base Model's Updated Architecture with Data Augmentation", size=15)
plt.show()
print("Preicision score:", precision_score(test_labels, predicted_lables, average="weighted"))
print("Recall score:", recall_score(test_labels, predicted_lables, average = "weighted"))
print("F1_score:", f1_score(test_labels, predicted_lables, average = "weighted"))
Preicision score: 0.5854724080172969 Recall score: 0.5131210191082802 F1_score: 0.5034306437499216
# importing the test datest again, so that this time images can be shuffled
# so that displayed images are not ordered in the same way as in the dataset
# and variety of classes can be examined
test_data_shuffled = tf.keras.utils.image_dataset_from_directory(test_dir, shuffle = True, seed = 247)
Found 3925 files belonging to 10 classes.
def right_format_image(pic):
'''
This function returns a
reshaped image into 224x224
format in terms of height and
width.
Further it normalizes the
pixel values within the range
of [0, 1].
'''
img_size = (224, 224)
image = tf.image.resize(pic, img_size)
image_expanded = np.expand_dims(image, axis=0)
image_copy = np.copy(image_expanded)
normalized = image_copy/255.
return normalized
def data_iterator(data):
'''
This function returns as arrays the
components of a batch.
'''
iterator = data.as_numpy_iterator()
batch = iterator.next()
return batch
# plotting images from the test dataset, with their actual and predicted from the model labels
predicted_batch = data_iterator(test_data_shuffled)
plt.figure(figsize=(12, 12))
plt.subplots_adjust(hspace = .1, wspace=.3)
plt.suptitle("Base Model's Updated Architecture with Data Augmentation", size = 20)
for i in range(9):
image, label = predicted_batch[0][i], predicted_batch[1][i]
predictions = model_adapted_augmented.predict(right_format_image(image))
prediction_label = class_names[predictions.argmax()]
ax = plt.subplot(3, 3, i + 1)
plt.imshow(image.astype(np.uint8))
plt.title("Actual label:{};\nPredicted label:{}".format(class_names[label],
class_names[predictions.argmax()]), size = 9)
plt.axis("off")
1/1 [==============================] - 0s 113ms/step 1/1 [==============================] - 0s 10ms/step 1/1 [==============================] - 0s 8ms/step 1/1 [==============================] - 0s 8ms/step 1/1 [==============================] - 0s 7ms/step
2023-03-17 12:00:35.460283: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.
1/1 [==============================] - 0s 9ms/step 1/1 [==============================] - 0s 7ms/step 1/1 [==============================] - 0s 8ms/step 1/1 [==============================] - 0s 8ms/step